• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD to prioritise flagship CPUs over lower end offerings

Amazing how companies bringing a fat wad of cash and research funding to the table can order products but johnny public can't get AMD to design and manufacture products at a loss with wishful thinking.
 
Nvidia actually said even with the additional funds consoles were lower margin. If you look at when AMD separated out it's financial metrics better - consoles were always lower. Zen R and D wasn't paid for by Sony or MS - it was paid for by AMD.

Also Intel had no consoles to subsidise it's CPU r and d and the billions it's lost on developing new nodes. New node development costs tens of billions of USD. Yet Intel client still makes a ton of money and decent margins. Nvidia makes decent machine yet you can find RTX3060 GPUs in £900 laptops,GTX1660 GPUs in loads of products. It seems both Intel and Nvidia spend more than AMD in R and D yet can supply both ends of the market.
AMD also has found $4 billion to buy back stocks. This is what Intel did and their new CEO said Intel wasted money on stock buybacks.

That is $4 billion they could have used for R and D or something else.

The reality is that Nvidia/Intel have increased marketshare over the last year. Nvidia makes more money than AMD does,and you can get entry level GPUs in prebuilt systems easily. Intel still makes more sales than AMD and you can find more them at all levels of the market. It's fake news to say mainstream products are loss making as Nvidia and Intel sell loads of OEM mainstream products and out revenue AMD. DIY buyers will always pay more than an OEM.

Jonny Public are still buying mostly Nvidia/Intel products and make more money than AMD does. This is because AMD has massive holes in its lineups and can't supply OEMs with enough volume.

80% of your 7nm wafer supply for the last 6 months towards consoles,when there is such massive demands for GPUs and CPUs,means AMD might have left billions on the table in terms of revenue. That is billions that Nvidia and Intel got in return.
 
Last edited:
Yields were apparently at under 60% so it makes me think they overpromised Sony/MS and had to push more wafers towards consoles to get the numbers expected. So this came at the expense of their consumer products,which they then jacked up in price.
While it is unlikely we'll ever know unless a reliable leaker shows up, my assumption is still that most of that 60% figure is from Sony.
And that Sony panicked once they found out the spec MS had gone for and raised clocks way past what the SOC was designed for - whether by themselves or after asking AMD what was possible.
The contracts are of course top secret, but it would be a monumentally bad contract if AMD were liable for Sony changing their minds. That is if they only get paid for functional dies at the new spec.
My speculation since hearing about Sony yields and the last minute push to clocks has always been that Sony must by now have millions of perfectly good, functional dies which cannot be binned to meet the PS5 specs.
Of course, even if AMD are being paid for the dies Sony cannot use, the consequence of all this more wafers going to console.
So in either case AMD are paying for it: either directly because Sony's contract is very one-sided and Sony don't pay for the wastage, or because AMD are not able to use those extra wafers for more profitable things (and let's be honest, even a Zen2 die cut down to 4C Athlon would probably be profitable than console margins).
As for Sony & MS underwriting AMD's R&D budget: well I think a lot of RNDA2 was paid for by them but since the console volumes prevent AMD from having any wafers to throw at RDNA2 cards it is all rather academic.
 
AMD also has found $4 billion to buy back stocks. This is what Intel did and their new CEO said Intel wasted money on stock buybacks.

Yet they've given no deadline, or indeed commitment, only saying up to $4B, which for AMD is a sensible move. They aren't as big as Intel as during the troublesome years the stock was heavily diluted in order to fund the R&D for Zen and to keep the company operating with pretty large debts. So they haven't 'found' $4B they've stated that is how much they will spend as a maximum, over no certain timescale, it could be 1 year or 5 years.

Saying something is a bad idea unless you fully understand the reason for doing it, well that is a bad idea, it raises investor confidence and shows confidence in your increasing margins, and market share and increase value to the investors who helped them when they were on the ropes.
 
While it is unlikely we'll ever know unless a reliable leaker shows up, my assumption is still that most of that 60% figure is from Sony.
And that Sony panicked once they found out the spec MS had gone for and raised clocks way past what the SOC was designed for - whether by themselves or after asking AMD what was possible.
The contracts are of course top secret, but it would be a monumentally bad contract if AMD were liable for Sony changing their minds. That is if they only get paid for functional dies at the new spec.
My speculation since hearing about Sony yields and the last minute push to clocks has always been that Sony must by now have millions of perfectly good, functional dies which cannot be binned to meet the PS5 specs.
Of course, even if AMD are being paid for the dies Sony cannot use, the consequence of all this more wafers going to console.
So in either case AMD are paying for it: either directly because Sony's contract is very one-sided and Sony don't pay for the wastage, or because AMD are not able to use those extra wafers for more profitable things (and let's be honest, even a Zen2 die cut down to 4C Athlon would probably be profitable than console margins).
As for Sony & MS underwriting AMD's R&D budget: well I think a lot of RNDA2 was paid for by them but since the console volumes prevent AMD from having any wafers to throw at RDNA2 cards it is all rather academic.
It's terrible planning from AMD. They got more allocations because of Huawei and used a lot of it for consoles. 200k wafers for consoles and a fraction of that for everything else.

All the people here laughed at Nvidia using an immature and worse Samsung 8nm node. Yet despite having larger GPUs which needed more power and cooling,JHH again succeeded somehow.

At least half their $5 billion last quarter was consumer graphics. AMD made $3.45 billion last quarter in total. Nvidia GPU client probably made more than all of AMD's CPU/GPU client business. Of that semi-custom made only $1.35 billion,and that included both server and console. So over 70% of AMD's 7NM wafer supply,derived less revenue than client with a fraction of the 7NM wafers.

That is how much money AMD has left on the table last year - it wouldn't surprise me one bit of they had more client supply,they would have probably matched or beaten Nvidia revenue.

Yet they've given no deadline, or indeed commitment, only saying up to $4B, which for AMD is a sensible move. They aren't as big as Intel as during the troublesome years the stock was heavily diluted in order to fund the R&D for Zen and to keep the company operating with pretty large debts. So they haven't 'found' $4B they've stated that is how much they will spend as a maximum, over no certain timescale, it could be 1 year or 5 years.

Saying something is a bad idea unless you fully understand the reason for doing it, well that is a bad idea, it raises investor confidence and shows confidence in your increasing margins, and market share and increase value to the investors who helped them when they were on the ropes.

I understand it fully as Intel did the same and their new CEO admitted it was a waste of money,which they could have spent on other areas of the company. Its more nonsense short termism which won't help them at all longterm. Many here defended Intel's buybacks as they made a tidy sum out of it,but in the end its a short term bandaid. Longterm the same investors would have simply made more money if Intel had invested more in its products.

Many people can't see AMD isn't out of the doldrums yet. Intel despite having so many problems massively outsells AMD,and Nvidia massively outsells AMD too. They did the same stuff during the Athlon 64 era by wasting precious funds,and then had no answer when their competitors woke up.

They should be using that money to secure more 5NM wafers,etc. I think you don't appreciate Intel is trying to buy as many TSMC wafers for 5NM/6NM as they can. So is Nvidia. They can easily outbid AMD. They realised what the big weakness of AMD is now.

AMD got very lucky with TSMC 7NM. Huawei was booted off TSMC due to the sanctions,so AMD got much more supply than they expected. Nvidia decided to experiment with Samsung. Whether this will play out with TSMC 5NM/6NM is another question.AMD is going to find it much more harder than with 7NM.

AMD can't expand revenue if they can't get enough supply.

Then investors will show less confidence in the company. Buy backs only work short term. They have literally told their investors their revenue growth is going to be limited by supply. That is not a good thing at all. Longterm investors care more about that.


Both Intel and Nvidia know this - if they can't beat them in technology they will use dirty tricks. The "new" Intel CEO was the same guy who was in when Intel did all its dirty tricks last time - he was one of the senior guys in their client and server computing section! This is because Intel could leverage better supply over the OEMs. AMD hit the same supply issues during the Athlon64 era.
 
Last edited:
It's terrible planning from AMD. They got more allocations because of Huawei and used a lot of it for consoles. 200k wafers for consoles and a fraction of that for everything else.

Can you prove how many wafers they have used, and what their allocation is, because it looks to me like you are just making up a random number.
 
80% of your 7nm wafer supply for the last 6 months towards consoles,when there is such massive demands for GPUs and CPUs,means AMD might have left billions on the table in terms of revenue. That is billions that Nvidia and Intel got in return.

You keep saying this like it's a casual choice by AMD.

The agreements for consoles go back a lot further than the current shortages and I don't recall you with a crystal ball back then saying don't get into the console contracts.
 
Can you prove how many wafers they have used, and what their allocation is, because it looks to me like you are just making up a random number.

Look in the thread - lots of links from Taiwanese industry sources covering TSMC. During the last quarter of 2020,AMD had an allocation of 150K 7NM wafers from TSMC,and used 120K for consoles.
This year they used 90K wafers for consoles. Its all there. So no unfortunately it's not a random number - its been linked by people such as KompuKare here before over the last year.

Surprised you missed it all.

Plus I think you have also not realised how big the console SOCs were. They are between 300MM2~360MM2 and will yield lower than any CPU. Even looking at the millions of consoles sold and back calculating stuff its still a huge number of 7NM wafers.

The semi-custom numbers are also inflated by the server side of the business which is very high margin. So that is still a ton of 7NM wafers for a relatively low revenue part of the business.

AMD has enough 7NM supply,but the consoles have made certain they are supply limited.


You keep saying this like it's a casual choice by AMD.

The agreements for consoles go back a lot further than the current shortages and I don't recall you with a crystal ball back then saying don't get into the console contracts.

AMD never expected so much 7NM allocations. Some of you have entirely forgotten what happened with Huawei. They were the 2nd largest 7NM partner with TSMC.

Taiwanese sources showed AMD had increased wafer allocations last year to consoles,and again you have forgotten Huawei was suddenly booted off TSMC too. Last year,AMD increased wafer allocations to consoles according to sources in Taiwan
 
Last edited:
Look in the thread - lots of links from Taiwanese industry sources covering TSMC. During the last quarter of 2020,AMD had an allocation of 150K 7NM wafers from TSMC,and used 120K for consoles.
This year they used 90K wafers for consoles. Its all there. So no unfortunately it's not a random number - its been linked by people such as KompuKare here before over the last year.

Taiwanese sources showed AMD had increased wafer allocations last year,and again you have forgotten Huawei was suddenly booted off TSMC too. Last year,AMD increase wafer allocations to consoles.
What do the contracts with Sony and Microsoft say?

(Rhetorical question ... not seriously expecting an answer.)

I have this feeling that AMD is not in a position to tell the console agreements to **** off so they can produce more profitable hardware at the current time.
 
Look in the thread - lots of links from Taiwanese industry sources covering TSMC. During the last quarter of 2020,AMD had an allocation of 150K 7NM wafers from TSMC,and used 120K for consoles.
This year they used 90K wafers for consoles. Its all there. So no unfortunately it's not a random number - its been linked by people such as KompuKare here before over the last year.

Surprised you missed it all.

So the answer was no, you can't prove it. It's random numbers in random articles with no actual evidence as TSMC doesn't break down how many wafers they supply to each customer.

So I didn't miss it, as it's all just made up.
 
They should be using that money to secure more 5NM wafers,etc. I think you don't appreciate Intel is trying to buy as many TSMC wafers for 5NM/6NM as they can. So is Nvidia. They can easily outbid AMD. They realised what the big weakness of AMD is now.

Just to highlight something, why would you pay scalped prices from TSMC which might increase your supply, but reduces your margins to the point where you may be making zero profit on the wafers once converted into chips and sold? Just so their EoQ results look worse as they've lowered EPS, and gross margins, that isn't smart and will negatively impact the business, even if they added a point or two to the MS for that quarter, or half.

Also type your whole response then post it, don't go back and add in like four paragraphs after, it's really hard to follow since when you go pretty much most of what you have written has been altered, and if you must do it, put it at the bottom and highlight it with "EDIT:" or just make another post.
 
I have this feeling that AMD is not in a position to tell the console agreements to **** off so they can produce more profitable hardware at the current time.

But they are in a position to not launch 3 different consoles,two new CPU lines and a new GPU line in one go,and then apparently wind down their older 12NM/14NM products at the same time. They really needed to have a backup plan of some sort - its quite clear Nvidia kept its 12NM/14NM GPU production lines in place,hence why you can find them everywhere in OEM prebuilt systems.


So the answer was no, you can't prove it. It's random numbers in random articles with no actual evidence as TSMC doesn't break down how many wafers they supply to each customer.

So I didn't miss it, as it's all just made up.

The issue is those sources have tended to be correct. The problem even if you are skeptical and ignore them,just look at how big console SOCs are,and back calculate the number of wafers required which some have tried. Then look at some of the market research companies which have good estimates of AMD CPU and GPU sales,and AMD/Nvidia/Intel uses these companies figures during investor meetings. This is the issue here - its quite clear semi-custom is much lower revenue overall,and that is inflated by the hugely profitable server CPUs.

They have allocated a huge amount of supply,even as a percentage of what TSMC does make each month. So far at least 14 million PS5 and XBox Series X consoles over two quarters IIRC,plus the XBox Series S. So 14 million 300~360MM2 SOCs. These are the size of something like a Navi10 GPU,or close to that of the one in the RTX3070. Or around 4X to 5X the size of a single Zen2/Zen3 CPU chiplet.

To put it in context,AMD sold 1.9 million GPUs in Q4 2020,and Nvidia just under 10 million(JPR and AMD/Nvidia use their figures). When it comes to Zen3 sales,apparently AMD sold close to 1 million Zen3 CPUs during Q4 2020(Mercury Research which again AMD uses).

Considering both Nvidia and Intel still massively outsell AMD,and AMD technologically wise is as good or better than its competitors,they have huge potential to take share from them. But not if they can't allocate enough volume because they have trapped themselves in overpromising too many consoles.

Nvidia/Intel have marketshare to lose.

The other issue is also pandemic spending. The last year has seen many people pull purchases forward,so how long is this sudden boom going to last? Then OFC you have mining too.

Edit!!

When the first AMD consoles all the news at the time was saying AMD supplied the finished chips to MS/Sony at an agreed price and had to eat any yield issues. This is why Nvidia had issues with MS over the original XBox. So is AMD moving to a licensing model then? Because I have not seen any indications so far.

So AMD has to pre- purchase wafer allocations and Sony/MS will pay for the finished products if that is that case. So the console contracts do affect what AMD can pre-purchase for their own products, because it reduces any remaining volume for everything else.

Just to highlight something, why would you pay scalped prices from TSMC which might increase your supply, but reduces your margins to the point where you may be making zero profit on the wafers once converted into chips and sold? Just so their EoQ results look worse as they've lowered EPS, and gross margins, that isn't smart and will negatively impact the business, even if they added a point or two to the MS for that quarter, or half.

Also type your whole response then post it, don't go back and add in like four paragraphs after, it's really hard to follow since when you go pretty much most of what you have written has been altered, and if you must do it, put it at the bottom and highlight it with "EDIT:" or just make another post.

The issue of AMD falling behind on nodes,is what eventually made Intel and Nvidia screw them over the last time,after AMD/ATI had managed to beat their competitors.

The issue here is what happens if AMD gets less 5NM/6NM supply relative to 7NM supply today?? At least AMD can target higher margin products like server CPUs,commercial GPUs,etc if they have supply and it costs more.This will at least relax constraints for the lagging nodes which can be used for volume products,where they can target more areas. If not you are back to square one,with everything needing one node and hence affecting supply.

If they don't,it not only means their competitors have more supply,they also have more supply of a better node.

AMD found every penny it could muster to get in early as possible on 7NM. They really need to make sure Nvidia/Intel don't sneak in. It was Nvidia sneaking in which eventually screwed AMD RTG/ATI over. ATI being consistently one node ahead of Nvidia helped a lot,but once they didn't have the lead,Nvidia took them to the cleaners.

Companies like Intel did contrarevenue and gave billions of USD away in bribes to deny AMD sales. The issue here is Intel/Nvidia most likely are going to try and do the same,if they know they can screw AMD over.
 
Last edited:
Saying something is a bad idea unless you fully understand the reason for doing it, well that is a bad idea, it raises investor confidence and shows confidence in your increasing margins, and market share and increase value to the investors who helped them when they were on the ropes.
However, what we are mainly debating here in our armchair CEO roles is that they seem to have sacrifices market share for margins.
Since TSMC at least don't have spare capacity, the sacrifice seems to have mainly come because of wafers to AMD's lowest margin area: consoles
 
So the answer was no, you can't prove it. It's random numbers in random articles with no actual evidence as TSMC doesn't break down how many wafers they supply to each customer.

So I didn't miss it, as it's all just made up.

It's definitely all a "trade secret", however we can make some educated guesses.
200k does seem a bit more than I got with my spreadsheet but then it is not over the same time frame.
aNtv5Kv.png

Anyway, as the last time I posted this, here's a Ethercalc link in case someone wants to play with it.
https://ethercalc.net/1vrvw3ho7ytq
Of course everything is speculation.
The PS5 yields are likely a lot worse than a theoretical 0.09 per mm defect rate would imply.
 
Of course everything is speculation.

So basically the entire discussion is almost pointless, as everything is speculation with little to no empirical evidence, the only thing remotely correct could be the die size.

I understand why people are interested, but there is little point with out hard facts, as it is just literally he said, she said.
 
So basically the entire discussion is almost pointless, as everything is speculation with little to no empirical evidence, the only thing remotely correct could be the die size.

I understand why people are interested, but there is little point with out hard facts, as it is just literally he said, she said.
Well it's a bit more than just one metric.
  1. die sizes
  2. some indication of defects per mm²
  3. Sales figures
    1. AMD supplied that 1,000,000 Zen 3 figure
    2. Sony keep boasting about their PS5 sales
So even ignoring defects and yields, we can come up with some idea of the wafers used and roughly what percentage of AMD's* slice of TMSC wafers is going to consoles.
* the slice might indeed be AMD's and Sony's and Microsoft's. We simply don't know who the contract works.
 
Well it's a bit more than just one metric.
  1. die sizes
  2. some indication of defects per mm²
  3. Sales figures
    1. AMD supplied that 1,000,000 Zen 3 figure
    2. Sony keep boasting about their PS5 sales
So even ignoring defects and yields, we can come up with some idea of the wafers used and roughly what percentage of AMD's* slice of TMSC wafers is going to consoles.
* the slice might indeed be AMD's and Sony's and Microsoft's. We simply don't know who the contract works.

But without knowing how many wafers AMD have then it isn't really data, it is just a guess. TSMC have said then predict to complete 1.2 million+ N7 wafers in 2021, and AMD are the largest N7 customer by far, so how many of the 1.2 million will go to AMD, you can guess since that is all you have been doing so far, 500k? 300k, more or less?

No one seems to be accounting for unshipped product, just assuming all wafers/silicon used have been allocated and shipped already, there was a delay in shipping Milan, yet servers are freely available now, meaning CPU stocks must be plentiful, there is no shortage of Zen3 based CPU's now other than the newer laptop and desktops APU parts but they seems to be coming though thick and fast.

By your reckoning if Sony shipped 20 million PS5's in 2021, that would take 155k wafers, so are you saying that would be a majority of the total wafers AMD have available? Add in the Xbox(s) at circa 10 million, and you have another 98k wafers, so in total over 250k wafers.

So what allocation would AMD have if they hadn't won the console bids many years ago when none of this shortage was predicted, and when the wafer supply agreements were signed? Do you honestly thiny they would have bought as many N7 wafers, would they even be in that position to afford it given the base level income that the console chips offer.

People banging on about AMD using too many wafers in consoles is just pointless, they wouldn't have the wafers if they didn't have the contracts, so they couldn't assigned them elsewhere anyhow.
 
Its complacency - they think Intel is finished for the immediate future,and think because they caught up with Nvidia,they can relax. However,Nvidia still made sure its "lower margin" 14NM/12NM products are in sufficient volume to equip most prebuild desktops/laptops with entry level GPUs. They have enough Ampere volume to sell RTX3060 GPUs in sub £1000 systems.

So Nvidia seems to have kept its eye on the ball,but AMD because its tasted success,has done another Athlon 64 and decided they won the premium brand battle. However,the premium brands like Nvidia have made sure they can supply OEM systems to the extent,you can't even find a 12NM/14NM AMD GPU anywhere in a prebuilt system.

Nvidia is laughing at both ends of the market. Plus you have at least a chance of getting an Ampere GPU at RRP. AMD can't even do that.



Nobody forced AMD to launch 3 consoles,Zen3 and RDNA2 at the same time.

AMD has pushed over 200000 wafers towards consoles if these reports are correct:
https://translate.google.com/translate?sl=auto&tl=en&u=https://ctee.com.tw/news/tech/375133.html
https://www.hardwareheaven.com/amd-80-of-production-reserved-for-consoles/
https://www.reddit.com/r/AMD_Stock/comments/iwxuhe/amd_placed_further_console_soc_order_at_tsmc/

They got extra capacity from Huawei being booted off TSMC too.Only a relatively small fraction of that went for othe products. They even increased wafer allocations to consoles late last year. Yields were apparently at under 60% so it makes me think they overpromised Sony/MS and had to push more wafers towards consoles to get the numbers expected. So this came at the expense of their consumer products,which they then jacked up in price.

Once you take away Epyc,etc out of the equation no wonder so many Zen3 chips took months to be easily available.

Just imagine if they had allocated half the number of wafers to consoles?? They would actually have had a full 7NM CPU/GPU stack and put Intel and Nvidia into serious problems.


Edit
 
Looking at it with 20:20 hindsight seems to be doing a bit of a disservice to their planning really...

When they planned Consoles/RDNA2/Zen3 around the same time they would've had no knowledge of the huge increase in demand, and accompanying supply issues, that are affecting the market today.

The deals with Sony/MS may be low margin but it's reliable income and the deals clearly part-funded RDNA2 development, keeping them happy now means they'll likely get the deal for PS6/Xbox <whatever> which is all good.

Yes they've favoured the Consoles, and consumer Zen3 market, with their wafer allocations but those are the markets that are the most reliable for them. OEMs have a reputation for being bought by Intel, and the consumer GPU market favours nvidia heavily. If nvidia could supply enough cards the number of RDNA2 cards sold even if also available would be relatively tiny, sure in hindsight nvidia couldn't/can't supply the market so there was an opening but that wasn't predictable.

They're being criticised for focussing on the low margin consoles and instead they should focus on <£250 CPU and OEM markets, which are low margin?

I doubt that the current focus will continue as/when the shortages (or supply) reduces, but for now the plan hasn't been terrible and it's a bit late at this stage to change it drastically, especially without ruining the relationship with Sony/MS which is one worth keeping (imo).
 
Back
Top Bottom